Bump actix-codec version, Reexport ConnectionIo

[brockelmore]

PR Type

Other

PR Checklist

• Tests for the changes have been added / updated.
• Documentation comments have been added / updated.
• A changelog entry has been made for the appropriate packages.
• Format code with the latest stable rustfmt.
• (Team) Label with affected crates and semver status.

Overview

During the refactor of moving actix_http::client over to awc, ConnectionIo was not reexported. Users need this trait for websocket usage. Additionally, updates the actix-codec version so websocket clients don't silently drop/not send continuation frames when the data is large (see actix/actix#431 and actix/actix-net#409)

[fakeshadow]

Can you show me how would you need the ConnectionIo trait? It's never intended to be exported as it's just a trait alias for writing bound easier.

[brockelmore]

Ah - I see i should use BoxedSocket. Previously, I used it as so:

/// A wrapper around a raw Websocket writer
pub type WsWriterRaw = SplitSink<Framed<Box<dyn ConnectionIo>, awc::ws::Codec>, awc::ws::Message>;

/// A wrapper around a raw Websocket stream
pub type WsStreamRaw = SplitStream<Framed<Box<dyn ConnectionIo>, awc::ws::Codec>>;

// ...
fn() -> Result<(WsWriterRaw, WsStreamRaw), ()> {
    let (_response, framed) = client_clone
        .ws(url)
        .connect()
        .await
        .map_err(|e| {
            tracing::error!("Websocket connection error: {}", e);
        })?;

    let (sink, stream) = framed.split();
    Ok((sink, stream))
}

But that can be replaced by BoxSocket i believe? But when i originally wrote the above function, the compiler said to use ConnectionIo in older versions of awc & actix-http

[fakeshadow]

yea I believe you can use BoxedSocket. It's a shortcut explicitly exported.
You can also write it in a generic way as Framed<BoxedSocket, awc::ws::Codec> can be described as S: Stream + Sink<Message> or impl Stream + Sink<Message> where Stream is used to receive and Sink is used to send message This way you don't have to import the concrete types and keep the code cleaner since the only type that is awc specific is Message, Frame and maybe ProtocolError.

[fakeshadow]

As for the compiler error I believe we should do better to hide the internal traits and types to reduce confusion.

[brockelmore]

gonna close this and allow someone else to bump codec versions whenever they want to release a new version